Underwater quadrotor vehicles operating in complex nearshore environments face challenges in three-dimensional path planning, obstacle-triggered replanning, and trajectory tracking under external disturbances. To address these issues, this paper evaluates an enhanced rapidly exploring random tree (iRRT) planner and a model predictive control–proportional–integral–derivative (MPC–PID) tracking strategy with residual disturbance estimation. The planner combines guided sampling, environment-aware adaptive extension, collision-checked reverse greedy pruning, and cubic B-spline smoothing. A bisection-based handoff reconnects the remaining safe path after a newly introduced obstacle; post-solution informed sampling is examined separately as an ablation. The tracking controller is evaluated on a nonlinear six-degree-of-freedom vehicle model under a seeded, wave-shaped disturbance proxy. In 30 runs on the original voxel map, both RRT and iRRT produced safe paths in the static and fixed-obstacle replanning cases under the same additional radius-two voxel-neighborhood test. Relative to RRT, iRRT reduced the mean number of raw path nodes by 37.4% in the static case and 12.8% after replanning, whereas its small mean path-length differences were not conclusive. Under matched paths and disturbance phases, adding residual estimation to MPC–PID reduced mean position RMSE by 12.1% on the static path and 9.7% on the replanned path. Component ablations and additional synthetic maps further examine the contributions and limits of the planning modules. These results support the proposed integration under the stated numerical assumptions; measured hydrodynamics and vehicle experiments remain to be investigated.